Implementing a White-Box Undetectable Backdoor for Random Fourier Features

DOI10.48550/arXiv.2609.16403MaRDI QIDQ7353394arXivFDO

Authors Samuel E. Jackson, Michael Collins, Jada Cumberland, Brianne Dunn, Ross Gore, Samuel Jackson, Sachin Shetty

Publication date 14 September 2026

Published in arXiv

Tldr We implement and test the white-box backdoor in RFF using commodity software tools, finding no evidence that it is detectable via weight-space or functional black-box comparisons, thereby confirming the feasibility of this threat model with standard scientific computing libraries.

Full work available at URL https://doi.org/10.48550/arXiv.2609.16403

arXiv classification

cs.CRcs.LG

Cites work

• An analysis of variance test for normality (complete samples)
• Computational Lower Bounds for Sparse PCA
• Continuous LWE
• Continuous LWE is as hard as LWE \& applications to learning Gaussian mixtures
• DISTRIBUTION OF EIGENVALUES FOR SOME SETS OF RANDOM MATRICES
• Lattice-based Cryptography
• On lattices, learning with errors, random linear codes, and cryptography
• Planting undetectable backdoors in machine learning models (extended abstract)
• Table for Estimating the Goodness of Fit of Empirical Distributions

This page was built for publication: Implementing a White-Box Undetectable Backdoor for Random Fourier Features

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7353394)